.env.example

# OpenAI-compatible model configuration (optional)
OPENAI_API_KEY=
OPENAI_BASE_URL=
MODEL_NAME=gpt-4o-mini

# Local storage
EDUPAPER_DATA_DIR=./data
CHROMA_DIR=./data/chroma
SQLITE_PATH=./data/edupaper.db
CHROMA_COLLECTION=education_papers

# Agent settings
ENABLE_ARXIV=true
ARXIV_MAX_RESULTS=5
LOCAL_SCORE_THRESHOLD=0.20
DEFAULT_TOP_K=6
MAX_CONTEXT_CHARS=16000
MAX_UPLOAD_MB=50

# PDF parsing: auto | pymupdf | mineru | paddleocr
# auto first tries native PyMuPDF text, then the fallback order below.
PDF_PARSER=auto
PDF_FALLBACK_ORDER=mineru,paddleocr
PDF_MIN_NATIVE_CHARS_PER_PAGE=80
PDF_MIN_NATIVE_PAGE_COVERAGE=0.50

# MinerU local CLI
MINERU_COMMAND=mineru
MINERU_BACKEND=pipeline
MINERU_TIMEOUT_SECONDS=900
# For restricted networks you may also set MINERU_MODEL_SOURCE=modelscope.

# PaddleOCR PP-StructureV3
PADDLEOCR_DEVICE=cpu
PADDLEOCR_USE_DOC_ORIENTATION_CLASSIFY=true
PADDLEOCR_USE_DOC_UNWARPING=false
PADDLEOCR_USE_TEXTLINE_ORIENTATION=true